The field of artificial intelligence has significantly evolved since the inception of Cyc, with the rise of neural networks and deep learning. These new methods, which rely on massive amounts of data and opaque algorithms, have been successful in solving previously intractable problems. However, they lack the common-sense knowledge that Cyc's rule-based, symbolic approach provides. The article suggests that Cyc, despite being seen as an anachronism, could still have a significant role to play in the future of AI, particularly in complementing large language models (LLMs).
Key takeaways:
- The Cyc project, initiated by the Cyc group at MCC, aimed to build a single intelligent agent with a knowledge base containing tens of millions of entries, with the goal of advancing work in expert systems, natural language understanding, and machine learning.
- The project, despite its significant achievements, has been largely forgotten due to the rise of new machine-learning algorithms that power modern technologies like chatbots and self-driving cars.
- Despite the shift in the field of artificial intelligence towards neural networks and deep learning, Cyc has continued to grow its knowledge base and remains funded by its commercial contracts.
- The late Doug Lenat, one of the key figures behind Cyc, believed that the system could complement large language models (LLMs), as Cyc's conclusions are always supported by a chain of reasoning that can be audited by a human, unlike LLMs which can be inconsistent and inaccurate.